Most Important Slide CS 594 Spring 2003 Lecture 1 Overview of High Performance Computing u Netlib software repository u Register for the na digest Go to http www netlib org Go to http www netlib org na net Register to receive the na digest http www netlib org na net join mail forw html Jack Dongarra Computer Science Department University of Tennessee 2 1 Computational Science u Why Turn to Simulation HPC offers a new way to do science u Experiment Theory Computation u Complex Large small Expensive Dangerous Computation used to approximate physical systems Advantages include Playing with simulation parameters to study emergent trends Possible replay of a particular simulation event Study systems where no exact theories exist When the problem is too u to do any other way Taurus to Taurus 60per 30deg mpeg 3 Pretty Pictures 4 Automotive Industry u Huge users of HPC technology u Main uses of simulation Ford is 25th largest user of HPC in the world u Main benefits 5 Aerodynamics similar to aerospace Crash simulation Metal sheet formation Noise vibration optimization Traffic simulation Reduced time to market of new cars Increased quality Reduced need to build prototypes more efficient integrated manufacturing processes 6 1 Units of High Performance Computing Why Turn to Simulation u u u Climate Weather Modeling Data intensive problems data mining oil reservoir simulation Problems with large length and time scales cosmology 1 Mflop s 1 Megaflop s 6 10 Flop sec 9 1 Gflop s 1 Gigaflop s 10 Flop sec 1 Tflop s 1 Teraflop s 10 Flop sec 1 Pflop s 1 Petaflop s 10 Flop sec 1 MB 1 Megabyte 10 Bytes 12 15 6 9 1 GB 1 Gigabyte 10 Bytes 1 TB 1 Terabyte 10 Bytes 1 PB 1 Petabyte 10 Bytes 12 15 7 8 High Performance Computing Today Technology Trends Microprocessor Capacity In the past decade the world has experienced one of the most exciting periods in computer development u Microprocessors have become smaller denser and more powerful u The result is that microprocessor based supercomputing is rapidly becoming the technology of preference in attacking some of the most important problems of science and engineering u Moore s Law 2X transistors Chip Every 1 5 years Called Moore s Law Microprocessors have become smaller denser and more powerful Not just processors bandwidth storage etc 9 4th Internet Revolution in Telecommunications u Telephone Radio Television u u u Traffic doubles every 100 days u u u 11 10 The Web Phenomenon u Growth in Internet outstrips the others u Exponential growth since 1985 u Gordon Moore co founder of Intel predicted in 1965 that the transistor density of semiconductor chips would double roughly every 18 months 90 93 Web invented U of Illinois Mosaic released March 94 0 1 traffic September 93 1 traffic w 200 sites June 94 10 of traffic w 2 000 sites Today 60 of traffic w 2 000 000 sites Every organization company school 12 2 Internet On Everything Peer to Peer Computing u u u u u Peer to peer is a style of networking in which a group of computers communicate directly with each other Wireless communication Home computer in the utility room next to the water heater and furnace Web tablets Imbedded computers in things all tied together Books furniture milk cartons etc u Smart Appliances Refrigerator scale etc 13 SETI home Global Distributed Computing u Running on 500 000 PCs 1000 CPU Years per Day 14 SETI home u 485 821 CPU Years so far u u Sophisticated Data Signal Processing Analysis Distributes Datasets from Arecibo Radio Telescope u u Use thousands of Internet connected PCs to help in the search for extraterrestrial intelligence When their computer is idle or being wasted this software will download a 300 kilobyte chunk of data for analysis Performs about 3 Tflops for each client in 15 hours The results of this analysis are sent back to the SETI team combined with thousands of other participants u Largest distributed computation project in existence Averaging 40 Tflop s u Today a number of companies trying this for profit 15 16 Grid Computing from ET toAnthrax u Google query attributes u Data centers 150M queries day 2000 second 3B documents in the index 15 000 Linux systems in 6 data centers 15 TFlop s and 1000 TB total capability 40 80 1U 2U servers cabinet 100 MB Ethernet switches cabinet with gigabit Ethernet uplink growth from 4 000 systems June 2000 18M queries then u 17 Performance and operation simple reissue of failed commands to new servers no performance debugging problems are not reproducible 18 Source Monika Henzinger Google 3 Performance vs Time Next Generation Web Performance VAX 780s 100 To treat CPU cycles and software like commodities Enable the coordinated use of geographically distributed resources in the absence of central control and existing trust relationships u Computing power is produced much like utilities such as power and water are produced for consumers u Users will have access to power on demand u This is one of our efforts at UT u u C RIS 10 yr ECL 15 60 4K Mips 65 Mhz yr Mips 25 mhz o 9000 uVAX 6K CMOS o 8600 TTL 1 0 780 5 Mhz 68K MIPS 8 Mhz MV10K C CIS uVAX OSr CMOS Will RISC continue on a CM y 60 x4 3 years 38 Moore s speed law 0 1 19 1980 1985 1990 Other Examples Sony PlayStation2 Trends in Computer Performance ASCI Red o u Emotion Engine 6 2 Gflop s 75 million polygons per second Microprocessor Report 13 5 Superscalar MIPS core vector coprocessor graphics DRAM Claim Toy Story realism brought to games About 250 21 22 Where Has This Performance Improvement Come From Sony PlayStation2 Export Limits u u u u u 23 Technology Organization Instruction Set Architecture Software Some combination of all of the above 24 4 1st Principles How fast can a serial computer be uWhat happens when the feature size shrinks by a factor of x uClock 1 Tflop 1 TB sequential machine u rate goes up by x actually less than x because of power consumption uTransistors uDie per unit area goes up by x 2 size also tends to increase typically another factor of x uRaw u computing power of the chip goes up by x4 of which x3 is devoted either to parallelism or locality r 3 mm Consider the 1 Tflop sequential machine data must travel some distance r to get from memory to CPU to get 1 data element per cycle this means 10 12 times per second at the speed of light c 3x108 m s so r c 1012 3 mm Now put 1 TB of storage in a 3 mm2 area each word occupies about 3 Angstroms2 the size of a small atom 26 Processor Memory Problem RAM Processors issue instructions roughly every nanosecond u
View Full Document
Unlocking...